Would it be a bad idea to move the static inline functions used in internal asserts to the header file?

[stduhpf]

My use case for this would be for applying LoRAs in sdcpp for example. If the weights don't have the expected shape, ggml_mul_mat()crashes the app with GGML_ASSERT(ggml_can_mul_mat(a, b)) failed. I think it would be nice to have acess to ggml_can_mul_mat() to handle this without crashing the whole thing. But maybe there's a good reaon I'm missing for why this is for internal use only. Any insight?

[ggerganov]

It's ok to expose those helper functions. Just expose those that you currently need for now.